for: 防止 Large Language Models (LLMs) 生成邪恶内容，包括骗财攻击。
methods: 使用四种常见的商业可用 LLMs（ChatGPT、GPT 4、Claude 和 Bard）生成功能骗财攻击，使用 serie 的邪恶提示。
results: 发现这些 LLMs 可以生成具有识别度的骗财电子邮件和网站，并且可以使用诸如逃脱检测系统的诡计来掩盖自己。Abstract
The advanced capabilities of Large Language Models (LLMs) have made them invaluable across various applications, from conversational agents and content creation to data analysis, research, and innovation. However, their effectiveness and accessibility also render them susceptible to abuse for generating malicious content, including phishing attacks. This study explores the potential of using four popular commercially available LLMs - ChatGPT (GPT 3.5 Turbo), GPT 4, Claude and Bard to generate functional phishing attacks using a series of malicious prompts. We discover that these LLMs can generate both phishing emails and websites that can convincingly imitate well-known brands, and also deploy a range of evasive tactics for the latter to elude detection mechanisms employed by anti-phishing systems. Notably, these attacks can be generated using unmodified, or "vanilla," versions of these LLMs, without requiring any prior adversarial exploits such as jailbreaking. As a countermeasure, we build a BERT based automated detection tool that can be used for the early detection of malicious prompts to prevent LLMs from generating phishing content attaining an accuracy of 97\% for phishing website prompts, and 94\% for phishing email prompts.
摘要
大型自然语言模型（LLM）的高级功能使得它们在不同应用程序中变得不可或缺，从对话代理和内容创作到数据分析、研究和创新。然而，它们的效iveness和可用性也使得它们容易遭受用于生成恶意内容的违用，包括骗财攻击。这项研究探讨了使用四种流行的商业可用的 LLM——ChatGPT（GPT 3.5 Turbo）、GPT 4、Claude 和 Bard——生成功能攻击。我们发现这些 LLM 可以生成具有识别度的恶意电子邮件和网站，并且可以部署一系列逃脱检测机制的诡计。值得注意的是，这些攻击可以使用未修改的、“纯净”的 LLM 进行生成，不需要任何先前的反对攻击如监禁。作为对策，我们构建了基于 BERT 的自动检测工具，可以用于早期检测恶意提示，以防止 LLM 生成攻击内容，其准确率为 97% для骗财网站提示，和 94% для骗财电子邮件提示。
Robustifying Language Models with Test-Time Adaptation
paper_authors: Noah Thomas McDermott, Junfeng Yang, Chengzhi Mao
for: 防止语言模型受到语言攻击
methods: 使用遮盖词预测来动态适应输入句子，以逆转语言攻击
results: 在两个受欢迎的句子分类任务上，我们的方法可以修复65%以上的语言攻击In English, this means:
for: Preventing language models from being attacked by adversarial language
methods: Using dynamic adaptation of input sentences with predictions from masked words to reverse language adversarial attacks
results: Our method can repair over 65% of adversarial language attacks on two popular sentence classification tasks without requiring any training.Abstract
Large-scale language models achieved state-of-the-art performance over a number of language tasks. However, they fail on adversarial language examples, which are sentences optimized to fool the language models but with similar semantic meanings for humans. While prior work focuses on making the language model robust at training time, retraining for robustness is often unrealistic for large-scale foundation models. Instead, we propose to make the language models robust at test time. By dynamically adapting the input sentence with predictions from masked words, we show that we can reverse many language adversarial attacks. Since our approach does not require any training, it works for novel tasks at test time and can adapt to novel adversarial corruptions. Visualizations and empirical results on two popular sentence classification datasets demonstrate that our method can repair adversarial language attacks over 65% o
摘要
大规模语言模型在多种语言任务上实现了状态机器的表现，但它们对语言攻击例子失败，这些例子是用来欺骗语言模型的，但对人类来说含义相同的句子。而现有的工作通常在训练时做 robustness 的优化，但对大规模基础模型来说，这种 retraining 是不现实的。因此，我们提议在测试时使用语言模型的 robustness。我们通过在输入句子上动态适应预测结果来显示，我们可以反转许多语言攻击例子。我们的方法不需要任何训练，因此它在测试时可以对新任务进行适应，并且可以适应新的语言攻击。我们的实验结果和视觉化结果表明，我们的方法可以修复大于 65% 的语言攻击例子。
Poisoning Retrieval Corpora by Injecting Adversarial Passages
results: 研究发现，这种攻击可以将紧密搜寻系统误导回答，并且这些陌生过程可以直接扩展到不同的域外查询和 corpora，例如在金融文档或网络论坛中，50个生成的过程可以误导>94%的查询。Abstract
Dense retrievers have achieved state-of-the-art performance in various information retrieval tasks, but to what extent can they be safely deployed in real-world applications? In this work, we propose a novel attack for dense retrieval systems in which a malicious user generates a small number of adversarial passages by perturbing discrete tokens to maximize similarity with a provided set of training queries. When these adversarial passages are inserted into a large retrieval corpus, we show that this attack is highly effective in fooling these systems to retrieve them for queries that were not seen by the attacker. More surprisingly, these adversarial passages can directly generalize to out-of-domain queries and corpora with a high success attack rate -- for instance, we find that 50 generated passages optimized on Natural Questions can mislead >94% of questions posed in financial documents or online forums. We also benchmark and compare a range of state-of-the-art dense retrievers, both unsupervised and supervised. Although different systems exhibit varying levels of vulnerability, we show they can all be successfully attacked by injecting up to 500 passages, a small fraction compared to a retrieval corpus of millions of passages.
摘要
dense retrievers 在多种信息检索任务中实现了状态码表现，但它们在实际应用中是否可以安全部署？在这项工作中，我们提出了一种 novel 的攻击方法，malicious user 通过修改精确的token来生成一小数量的对抗 passage，以最大化与提供的训练问题的相似性。当这些对抗 passage 添加到大量的检索库中时，我们发现这种攻击非常有效，能让这些系统 retrieve 这些恶意生成的 passage 作为未看过的查询。即使在不同的领域和数据集上，这些对抗 passage 仍然能够直接泛化，并达到高度的成功攻击率。例如，我们发现50个优化后的对抗 passage 可以误导 >94%的金融文档或在线讨论中的问题。我们还对多种当前最佳的 dense retrievers 进行了 benchmark 和比较，包括不超过500个对抗 passage 的攻击。虽然不同的系统在攻击上展现出不同的抵抗程度，但我们发现所有这些系统都可以被成功攻击。
BERT Lost Patience Won’t Be Robust to Adversarial Slowdown
results: 研究发现，钝化攻击可以减少多出口机制提供的计算成本，特别是对于复杂的机制而言。 此外，研究还发现了一些常见的 perturbation 模式，并与标准的 adversarial text 攻击进行比较。 最后，研究发现了输入整形可以有效地解决钝化攻击，但是 adversarial training 无法战胜钝化攻击。Abstract
In this paper, we systematically evaluate the robustness of multi-exit language models against adversarial slowdown. To audit their robustness, we design a slowdown attack that generates natural adversarial text bypassing early-exit points. We use the resulting WAFFLE attack as a vehicle to conduct a comprehensive evaluation of three multi-exit mechanisms with the GLUE benchmark against adversarial slowdown. We then show our attack significantly reduces the computational savings provided by the three methods in both white-box and black-box settings. The more complex a mechanism is, the more vulnerable it is to adversarial slowdown. We also perform a linguistic analysis of the perturbed text inputs, identifying common perturbation patterns that our attack generates, and comparing them with standard adversarial text attacks. Moreover, we show that adversarial training is ineffective in defeating our slowdown attack, but input sanitization with a conversational model, e.g., ChatGPT, can remove perturbations effectively. This result suggests that future work is needed for developing efficient yet robust multi-exit models. Our code is available at: https://github.com/ztcoalson/WAFFLE
摘要
在这篇论文中，我们系统地评估了多出口语言模型对针对性慢速攻击的Robustness。为了审计其Robustness，我们设计了一种通过绕过早出点生成自然针对性文本的攻击。我们使用这种WAFFLE攻击来进行对三种多出口机制的GLUEbenchmark进行广泛的评估，并显示我们的攻击可以在白盒和黑盒设置下减少了这些方法提供的计算成本。我们发现，复杂的机制更容易受到针对性慢速攻击。此外，我们还进行了文本输入的语言分析，找到了我们的攻击生成的扰乱模式，并与标准针对性攻击相比较。此外，我们还发现，对于我们的慢速攻击，反向训练无法有效地抵抗，但是使用 conversational 模型，例如 ChatGPT，可以有效地除掉扰乱。这种结果表明，未来的工作需要开发高效又Robust的多出口模型。我们的代码可以在：https://github.com/ztcoalson/WAFFLE 获取。
Learning to Follow Object-Centric Image Editing Instructions Faithfully
results: 对比于现有的基线，该方法能够进行细化的对象中心编辑，并且能够在未经训练的领域中进行扩展。此外，模型还能够捕捉到文本指令中的含义，进行 faithfulness 的捕捉和修改。Abstract
Natural language instructions are a powerful interface for editing the outputs of text-to-image diffusion models. However, several challenges need to be addressed: 1) underspecification (the need to model the implicit meaning of instructions) 2) grounding (the need to localize where the edit has to be performed), 3) faithfulness (the need to preserve the elements of the image not affected by the edit instruction). Current approaches focusing on image editing with natural language instructions rely on automatically generated paired data, which, as shown in our investigation, is noisy and sometimes nonsensical, exacerbating the above issues. Building on recent advances in segmentation, Chain-of-Thought prompting, and visual question answering, we significantly improve the quality of the paired data. In addition, we enhance the supervision signal by highlighting parts of the image that need to be changed by the instruction. The model fine-tuned on the improved data is capable of performing fine-grained object-centric edits better than state-of-the-art baselines, mitigating the problems outlined above, as shown by automatic and human evaluations. Moreover, our model is capable of generalizing to domains unseen during training, such as visual metaphors.
摘要
自然语言指令是文本到图像扩散模型的高级用户界面。然而，需要解决以下挑战：1）下pecification（需要模型理解指令的隐含含义）、2）grounding（需要确定编辑操作的具体位置）、3）loyal（需要保持图像中未受影响的元素）。现有的方法通过自动生成的对应数据来实现图像编辑，但这些数据经过我们的调查发现噪音和无意义，这些问题进一步加剧了上述问题。我们基于最近的分割、链条提示和视觉问答技术进行了大幅改进，提高对应数据的质量。此外，我们还强调要更改的图像部分，以提高模型的精细化对象编辑能力。经过练习这些改进后的模型，在自动和人工评估中都能够更好地完成细化的对象编辑任务，并且能够在训练时未看到的领域中进行推断。此外，我们的模型还能够捕捉到视觉 метаFOR，进一步提高图像编辑的精度和效果。
Women Wearing Lipstick: Measuring the Bias Between an Object and Its Related Gender
for: investigate the impact of objects on gender bias in image captioning systems
methods: use visual semantic-based gender score to measure the degree of bias
results: propose a gender score that can be used as an additional metric to existing approach, and observe the bias relation between caption and related genderAbstract
In this paper, we investigate the impact of objects on gender bias in image captioning systems. Our results show that only gender-specific objects have a strong gender bias (e.g., women-lipstick). In addition, we propose a visual semantic-based gender score that measures the degree of bias and can be used as a plug-in for any image captioning system. Our experiments demonstrate the utility of the gender score, since we observe that our score can measure the bias relation between a caption and its related gender; therefore, our score can be used as an additional metric to the existing Object Gender Co-Occ approach. Code and data are publicly available at \url{https://github.com/ahmedssabir/GenderScore}.
摘要
在这篇论文中，我们研究了图像描述系统中的性别偏见。我们的结果显示，只有性别特定的物品会带有强烈的性别偏见（例如女性 lipstick）。此外，我们提出了基于视觉 semantics 的性别分数，可以用于任何图像描述系统中。我们的实验表明了这个分数的用途，因为我们发现了这个分数可以测量描述和其相关的性别之间的偏见关系，因此可以用作现有的 Object Gender Co-Occ 方法的附加指标。代码和数据都可以在 \url{https://github.com/ahmedssabir/GenderScore} 上获取。
Unified Representation for Non-compositional and Compositional Expressions
for: This paper is written for researchers and developers working on natural language processing (NLP) and machine learning, specifically those interested in non-compositional language and idiomatic expressions.
methods: The paper proposes a language model called PIER, which builds on BART and generates semantically meaningful and contextually appropriate representations for English potentially idiomatic expressions (PIEs).
results: The paper shows that the representations generated by PIER result in higher homogeneity scores for embedding clustering and gains in accuracy and sequence accuracy for PIE sense classification and span detection compared to the state-of-the-art IE representation model, GIEA, without sacrificing performance on NLU tasks.Abstract
Accurate processing of non-compositional language relies on generating good representations for such expressions. In this work, we study the representation of language non-compositionality by proposing a language model, PIER, that builds on BART and can create semantically meaningful and contextually appropriate representations for English potentially idiomatic expressions (PIEs). PIEs are characterized by their non-compositionality and contextual ambiguity in their literal and idiomatic interpretations. Via intrinsic evaluation on embedding quality and extrinsic evaluation on PIE processing and NLU tasks, we show that representations generated by PIER result in 33% higher homogeneity score for embedding clustering than BART, whereas 3.12% and 3.29% gains in accuracy and sequence accuracy for PIE sense classification and span detection compared to the state-of-the-art IE representation model, GIEA. These gains are achieved without sacrificing PIER's performance on NLU tasks (+/- 1% accuracy) compared to BART.
摘要
Accurate processing of non-compositional language relies on generating good representations for such expressions. In this work, we study the representation of language non-compositionality by proposing a language model, PIER, that builds on BART and can create semantically meaningful and contextually appropriate representations for English potentially idiomatic expressions (PIEs). PIEs are characterized by their non-compositionality and contextual ambiguity in their literal and idiomatic interpretations. Via intrinsic evaluation on embedding quality and extrinsic evaluation on PIE processing and NLU tasks, we show that representations generated by PIER result in 33% higher homogeneity score for embedding clustering than BART, whereas 3.12% and 3.29% gains in accuracy and sequence accuracy for PIE sense classification and span detection compared to the state-of-the-art IE representation model, GIEA. These gains are achieved without sacrificing PIER's performance on NLU tasks (+/- 1% accuracy) compared to BART.Here's the translation in Traditional Chinese: Accurate processing of non-compositional language relies on generating good representations for such expressions. In this work, we study the representation of language non-compositionality by proposing a language model, PIER, that builds on BART and can create semantically meaningful and contextually appropriate representations for English potentially idiomatic expressions (PIEs). PIEs are characterized by their non-compositionality and contextual ambiguity in their literal and idiomatic interpretations. Via intrinsic evaluation on embedding quality and extrinsic evaluation on PIE processing and NLU tasks, we show that representations generated by PIER result in 33% higher homogeneity score for embedding clustering than BART, whereas 3.12% and 3.29% gains in accuracy and sequence accuracy for PIE sense classification and span detection compared to the state-of-the-art IE representation model, GIEA. These gains are achieved without sacrificing PIER's performance on NLU tasks (+/- 1% accuracy) compared to BART.
PACuna: Automated Fine-Tuning of Language Models for Particle Accelerators
results: PACuna可以解决复杂的加速器问题，并被专家 validateAbstract
Navigating the landscape of particle accelerators has become increasingly challenging with recent surges in contributions. These intricate devices challenge comprehension, even within individual facilities. To address this, we introduce PACuna, a fine-tuned language model refined through publicly available accelerator resources like conferences, pre-prints, and books. We automated data collection and question generation to minimize expert involvement and make the data publicly available. PACuna demonstrates proficiency in addressing intricate accelerator questions, validated by experts. Our approach shows adapting language models to scientific domains by fine-tuning technical texts and auto-generated corpora capturing the latest developments can further produce pre-trained models to answer some intricate questions that commercially available assistants cannot and can serve as intelligent assistants for individual facilities.
摘要
在加速器领域的探索中，由于最近的贡献增加， navigating 已成为越来越复杂的任务。这些细腻的设备会使人们感到困惑，甚至在同一个设施内。为解决这问题，我们介绍了 PACuna，一种精度调整的语言模型，通过公共可用的加速器资源，如会议、预印和书籍来优化。我们自动收集数据和生成问题，以最小化专家参与度，并将数据公开可用。 PACuna 在解决复杂的加速器问题方面表现出色，由专家 validate。我们的方法表明，通过科学领域中的技术文本和自动生成的 corpora 来练习语言模型，可以生成适用于一些复杂问题的预训练模型，这些模型可以作为加速器设施的智能助手。
Pushdown Layers: Encoding Recursive Structure in Transformer Language Models
paper_authors: Shikhar Murty, Pratyusha Sharma, Jacob Andreas, Christopher D. Manning
for: This paper aims to improve the syntactic generalization of Transformer language models by introducing a new self-attention layer called Pushdown Layers.
methods: The Pushdown Layers model recursive state via a stack tape that tracks estimated depths of every token, and the Transformer LMs with Pushdown Layers use this stack tape to softly modulate attention over tokens.
results: The authors achieve dramatically better and 3-5x more sample-efficient syntactic generalization when training Transformers equipped with Pushdown Layers on a corpus of strings annotated with silver constituency parses, while maintaining similar perplexities.Abstract
Recursion is a prominent feature of human language, and fundamentally challenging for self-attention due to the lack of an explicit recursive-state tracking mechanism. Consequently, Transformer language models poorly capture long-tail recursive structure and exhibit sample-inefficient syntactic generalization. This work introduces Pushdown Layers, a new self-attention layer that models recursive state via a stack tape that tracks estimated depths of every token in an incremental parse of the observed prefix. Transformer LMs with Pushdown Layers are syntactic language models that autoregressively and synchronously update this stack tape as they predict new tokens, in turn using the stack tape to softly modulate attention over tokens -- for instance, learning to "skip" over closed constituents. When trained on a corpus of strings annotated with silver constituency parses, Transformers equipped with Pushdown Layers achieve dramatically better and 3-5x more sample-efficient syntactic generalization, while maintaining similar perplexities. Pushdown Layers are a drop-in replacement for standard self-attention. We illustrate this by finetuning GPT2-medium with Pushdown Layers on an automatically parsed WikiText-103, leading to improvements on several GLUE text classification tasks.
摘要
人类语言中具有重要特点的一种是Recursion，它对于自注意机制的缺乏显式状态跟踪机制而具有挑战性。因此，Transformer语言模型在捕捉长尾递归结构方面表现不佳，并且 exhibit sample-inefficient sintactic generalization。这项工作介绍了Pushdown层，一种新的自注意层，通过一个堆栈带跟踪每个字符的估计深度来模型 recursive state。Transformer LMs WITH Pushdown Layers 是一种强式语言模型，可以同步和顺序地更新这个堆栈带，并在预测新字符时使用堆栈来软模式地修饰注意力。例如，学习"跳过"关闭的成分。当在一个Silver Constituency Parses 的集合上训练 Transformer 时，它们配备 Pushdown Layers 可以在同样的批量大小下达到更好的 3-5 倍的样本效率，同时保持相似的折衔率。Pushdown Layers 是一种可替换的自注意层。我们通过对 GPT2-medium WITH Pushdown Layers 在自动生成的 WikiText-103 上进行训练，来示例这一点。
A Survey on Recent Named Entity Recognition and Relation Classification Methods with Focus on Few-Shot Learning Approaches
results: 本文对两个领域的最新成果进行了比较分析，并对几步学习方法的结果进行了结构化分析。Abstract
Named entity recognition and relation classification are key stages for extracting information from unstructured text. Several natural language processing applications utilize the two tasks, such as information retrieval, knowledge graph construction and completion, question answering and other domain-specific applications, such as biomedical data mining. We present a survey of recent approaches in the two tasks with focus on few-shot learning approaches. Our work compares the main approaches followed in the two paradigms. Additionally, we report the latest metric scores in the two tasks with a structured analysis that considers the results in the few-shot learning scope.
摘要
Named entity recognition和关系分类是抽取无结构文本信息的关键阶段。许多自然语言处理应用程序利用这两个任务，如信息检索、知识图构建和完善、问答等领域应用程序，以及生物医学数据挖掘等领域应用程序。我们对最近的方法进行了评论，并对几种学习 paradigms进行了比较。此外，我们还对这两个任务中最新的 метри克分数进行了报告，并进行了结构化分析，考虑到几种少量学习范围内的结果。
ArBanking77: Intent Detection Neural Model and a New Dataset in Modern and Dialectical Arabic
results: 本研究实现了对 live chat 查询中的实际应用，并在 simulated low-resource 环境下进行了广泛的实验，以评估模型在不同的情况下的表现。Abstract
This paper presents the ArBanking77, a large Arabic dataset for intent detection in the banking domain. Our dataset was arabized and localized from the original English Banking77 dataset, which consists of 13,083 queries to ArBanking77 dataset with 31,404 queries in both Modern Standard Arabic (MSA) and Palestinian dialect, with each query classified into one of the 77 classes (intents). Furthermore, we present a neural model, based on AraBERT, fine-tuned on ArBanking77, which achieved an F1-score of 0.9209 and 0.8995 on MSA and Palestinian dialect, respectively. We performed extensive experimentation in which we simulated low-resource settings, where the model is trained on a subset of the data and augmented with noisy queries to simulate colloquial terms, mistakes and misspellings found in real NLP systems, especially live chat queries. The data and the models are publicly available at https://sina.birzeit.edu/arbanking77.
摘要
Note: "AraBERT" is a pre-trained Arabic language model, similar to BERT.
SALMA: Arabic Sense-Annotated Corpus and WSD Benchmarks
results: 研究人员通过了多种 metric（Kappa、Lineal Weighted Kappa、Quadratic Weighted Kappa、Mean Average Error、Root Mean Square Error）来评估注释质量，并发现了非常高的间接对应者一致性。此外，研究人员还开发了一个基于 Target Sense Verification 的 Word Sense Disambiguation 系统，并使用这个系统来评估三种 Target Sense Verification 模型的性能，其中最佳模型的准确率达到了 84.2%（使用 Modern）和 78.7%（使用 Ghani）。Abstract
SALMA, the first Arabic sense-annotated corpus, consists of ~34K tokens, which are all sense-annotated. The corpus is annotated using two different sense inventories simultaneously (Modern and Ghani). SALMA novelty lies in how tokens and senses are associated. Instead of linking a token to only one intended sense, SALMA links a token to multiple senses and provides a score to each sense. A smart web-based annotation tool was developed to support scoring multiple senses against a given word. In addition to sense annotations, we also annotated the corpus using six types of named entities. The quality of our annotations was assessed using various metrics (Kappa, Linear Weighted Kappa, Quadratic Weighted Kappa, Mean Average Error, and Root Mean Square Error), which show very high inter-annotator agreement. To establish a Word Sense Disambiguation baseline using our SALMA corpus, we developed an end-to-end Word Sense Disambiguation system using Target Sense Verification. We used this system to evaluate three Target Sense Verification models available in the literature. Our best model achieved an accuracy with 84.2% using Modern and 78.7% using Ghani. The full corpus and the annotation tool are open-source and publicly available at https://sina.birzeit.edu/salma/.
摘要
SALMA，首个阿拉伯语意义注释 корпу斯，包含约34000个字符，所有字符都有意义注释。 corpora 使用两个不同的意义集 simultaneously (Modern 和 Ghani)。 SALMA 的创新在于如何将字符和意义相关联。而不是将字符与一个固定的意义相关联，SALMA 将字符与多个意义相关联，并为每个意义提供一个分数。为支持多个意义对一个词的分数，我们开发了一个智能的网络基于的注释工具。此外，我们还对 corpora 进行了六种命名实体的注释。我们对注释的质量进行了多种 metric 评估（Kappa、线性权重Kappa、quadratic Weighted Kappa、平均误差和根平方误差），它们显示了非常高的间对注释者一致性。为建立基于我们 SALMA корпу的单词意义推断基线，我们开发了一个 Target Sense Verification 基于的全 End-to-end Word Sense Disambiguation 系统。我们使用这个系统来评估Literature 中提供的三种 Target Sense Verification 模型。我们的最佳模型在Modern 和 Ghani 中达到了84.2%和78.7%的准确率。整个corpus 和注释工具都是开源的，可以在 获取。
LLMs and Finetuning: Benchmarking cross-domain performance for hate speech detection
results: 我们的研究结果表明，通过适度调整和更多的标签多样性，可以提高模型的泛化性和检测精度。我们认为未来的仇恨言语检测应该强调cross-domain泛化和合适的benchmarking实践。Abstract
This paper compares different pre-trained and fine-tuned large language models (LLMs) for hate speech detection. Our research underscores challenges in LLMs' cross-domain validity and overfitting risks. Through evaluations, we highlight the need for fine-tuned models that grasp the nuances of hate speech through greater label heterogeneity. We conclude with a vision for the future of hate speech detection, emphasizing cross-domain generalizability and appropriate benchmarking practices.
摘要
这篇论文比较了不同的预训练和微调大型自然语言模型（LLM）对仇视言语检测的性能。我们的研究强调了LLM在不同领域的交叉领域有效性和过拟合风险。通过评估，我们强调需要微调模型，以便更好地捕捉仇视言语的细节和多样性。我们 conclude with a future vision for hate speech detection，强调跨领域一致性和合适的标准化实践。Note that the word " LL" in the original text was translated as "LLM" in Simplified Chinese, as "LL" is not a commonly used term in Simplified Chinese.
S2F-NER: Exploring Sequence-to-Forest Generation for Complex Entity Recognition
results: 我们的模型在三个不连续NER数据集和两个嵌入NER数据集上表现出色，特别是对于不连续实体识别。Abstract
Named Entity Recognition (NER) remains challenging due to the complex entities, like nested, overlapping, and discontinuous entities. Existing approaches, such as sequence-to-sequence (Seq2Seq) generation and span-based classification, have shown impressive performance on various NER subtasks, but they are difficult to scale to datasets with longer input text because of either exposure bias issue or inefficient computation. In this paper, we propose a novel Sequence-to-Forest generation paradigm, S2F-NER, which can directly extract entities in sentence via a Forest decoder that decode multiple entities in parallel rather than sequentially. Specifically, our model generate each path of each tree in forest autoregressively, where the maximum depth of each tree is three (which is the shortest feasible length for complex NER and is far smaller than the decoding length of Seq2Seq). Based on this novel paradigm, our model can elegantly mitigates the exposure bias problem and keep the simplicity of Seq2Seq. Experimental results show that our model significantly outperforms the baselines on three discontinuous NER datasets and on two nested NER datasets, especially for discontinuous entity recognition.
摘要
Retrofitting Light-weight Language Models for Emotions using Supervised Contrastive Learning
results: 这篇论文的结果显示，使用这种方法更新预训网络的BERT和RoBERTa模型，可以实现情感识别的改进。对于情感分析和讽刺检测任务，这些模型比预训网络原始版本（约1%的提升）和其他已知方法更好。此外，这些更新后的模型在少量学习设定下表现更好。Abstract
We present a novel retrofitting method to induce emotion aspects into pre-trained language models (PLMs) such as BERT and RoBERTa. Our method updates pre-trained network weights using contrastive learning so that the text fragments exhibiting similar emotions are encoded nearby in the representation space, and the fragments with different emotion content are pushed apart. While doing so, it also ensures that the linguistic knowledge already present in PLMs is not inadvertently perturbed. The language models retrofitted by our method, i.e., BERTEmo and RoBERTaEmo, produce emotion-aware text representations, as evaluated through different clustering and retrieval metrics. For the downstream tasks on sentiment analysis and sarcasm detection, they perform better than their pre-trained counterparts (about 1% improvement in F1-score) and other existing approaches. Additionally, a more significant boost in performance is observed for the retrofitted models over pre-trained ones in few-shot learning setting.
摘要
我们提出了一种新的改进方法，用于启用语言模型（PLM）中的情感方面，如BERT和RoBERTa。我们的方法通过对预训练网络权重进行更新，使得表达同样情感的文本片段在表示空间中相近，而表达不同情感的片段则被推迟。同时，我们的方法 также确保了预训练语言模型中的语言知识不会偶然受到影响。我们修改后的语言模型，即BERTEmo和RoBERTaEmo，可以生成情感意识的文本表示，根据不同的聚类和检索指标进行评估。在情感分析和讽刺检测下投入下，它们与预训练模型（大约1%的提升）和其他现有方法相比，表现出较好的性能。此外，我们发现在少量学习 Setting中，修改后的模型比预训练模型表现更好，具有更大的提升。
Sentence Bag Graph Formulation for Biomedical Distant Supervision Relation Extraction
results: 在两个大规模生物医学关系集和 NYT 集上进行了广泛的实验，并证明了我们提出的方法可以在生物医学数据中 distant supervision relation extraction 中表现出色，同时也在普通文本挖掘领域中表现出优秀的relation extraction 能力。Abstract
We introduce a novel graph-based framework for alleviating key challenges in distantly-supervised relation extraction and demonstrate its effectiveness in the challenging and important domain of biomedical data. Specifically, we propose a graph view of sentence bags referring to an entity pair, which enables message-passing based aggregation of information related to the entity pair over the sentence bag. The proposed framework alleviates the common problem of noisy labeling in distantly supervised relation extraction and also effectively incorporates inter-dependencies between sentences within a bag. Extensive experiments on two large-scale biomedical relation datasets and the widely utilized NYT dataset demonstrate that our proposed framework significantly outperforms the state-of-the-art methods for biomedical distant supervision relation extraction while also providing excellent performance for relation extraction in the general text mining domain.
摘要
我们提出了一种新的图structure-based框架，用于解决远程supervised关系抽取中的一些主要挑战，并在生物医学数据中进行了实质性的证明。特别是，我们提出了一种将句子袋视为实体对的图视图，使得对于实体对的信息在句子袋中进行消息传递基于的聚合。该提议的框架可以解决远程supervised关系抽取中的常见问题，即标签杂乱，并有效地 incorporate句子之间的依赖关系。我们在两个大规模的生物医学关系数据集和常用的NYT数据集上进行了广泛的实验，得到了我们提议的框架在生物医学关系抽取中的显著超越州方法的性能，同时也在文本挖掘领域中表现出色。
Pre-trained Speech Processing Models Contain Human-Like Biases that Propagate to Speech Emotion Recognition
paper_authors: Isaac Slaughter, Craig Greenberg, Reva Schwartz, Aylin Caliskan
for: 这个研究旨在检测语音处理模型中的偏见，具体来说是检测预训练模型中的偏见。
methods: 这个研究使用了Speech Embedding Association Test（SpEAT）来检测预训练模型中的偏见。SpEAT是基于自然语言处理中的词嵌入协会测试，用于量化模型对不同概念的偏见，如种族、性别等。
results: 这个研究发现了14种预训练模型中的偏见，包括abled人群对disabled人群的正面偏见、欧洲裔美国人群对非洲裔美国人群的正面偏见、女性对♂的正面偏见、美国口音 speaker对非美国口音 speaker的正面偏见、年轻人群对老年人群的正面偏见。此外，研究还发现了这些偏见在下游任务Speech Emotion Recognition（SER）中的影响。在66个测试中（69%），由SpEAT测试发现的偏见与SER任务中的偏见相关。Abstract
Previous work has established that a person's demographics and speech style affect how well speech processing models perform for them. But where does this bias come from? In this work, we present the Speech Embedding Association Test (SpEAT), a method for detecting bias in one type of model used for many speech tasks: pre-trained models. The SpEAT is inspired by word embedding association tests in natural language processing, which quantify intrinsic bias in a model's representations of different concepts, such as race or valence (something's pleasantness or unpleasantness) and capture the extent to which a model trained on large-scale socio-cultural data has learned human-like biases. Using the SpEAT, we test for six types of bias in 16 English speech models (including 4 models also trained on multilingual data), which come from the wav2vec 2.0, HuBERT, WavLM, and Whisper model families. We find that 14 or more models reveal positive valence (pleasantness) associations with abled people over disabled people, with European-Americans over African-Americans, with females over males, with U.S. accented speakers over non-U.S. accented speakers, and with younger people over older people. Beyond establishing that pre-trained speech models contain these biases, we also show that they can have real world effects. We compare biases found in pre-trained models to biases in downstream models adapted to the task of Speech Emotion Recognition (SER) and find that in 66 of the 96 tests performed (69%), the group that is more associated with positive valence as indicated by the SpEAT also tends to be predicted as speaking with higher valence by the downstream model. Our work provides evidence that, like text and image-based models, pre-trained speech based-models frequently learn human-like biases. Our work also shows that bias found in pre-trained models can propagate to the downstream task of SER.
摘要
先前的研究已经证明人的民族和语言风格会影响语音处理模型对他们的性能。但是这种偏见来自哪里？在这项工作中，我们介绍了语音嵌入协会测试（SpEAT），用于检测语音处理模型中的偏见。SpEAT Draws inspiration from natural language processing中的嵌入协会测试，用于衡量不同概念的嵌入表示，如种族或语言风格，并捕捉模型从大规模社会文化数据中学习的人类化偏见。使用SpEAT，我们测试了16种英语语音模型（包括4种多语言模型），来自wav2vec 2.0、HuBERT、WavLM和Whisper模型家族。我们发现14个或更多的模型表现出了有利可能（愉悦）偏见，即abled人群比 disabled人群更有利可能，European-Americans比 African-Americans更有利可能，女性比男性更有利可能，U.S.口音说话者比非U.S.口音说话者更有利可能，以及年轻人比老年人更有利可能。我们不仅证明了语音处理模型中的这些偏见，还表明它们可能有实际的影响。我们比较了预训练模型中的偏见和下游任务speech emotion recognition（SER）模型中的偏见，发现在96次测试中（69%），与预训练模型中的偏见相关的组 Also tends to be predicted as speaking with higher valence by the downstream model。我们的工作证明了，如文本和图像基于模型一样，预训练语音基于模型经常学习人类化偏见。我们的工作还表明了预训练模型中的偏见可能会传播到下游任务中。
MUST: A Multilingual Student-Teacher Learning approach for low-resource speech recognition
paper_authors: Muhammad Umar Farooq, Rehan Ahmad, Thomas Hain
for: 本研究旨在解决语音识别系统训练中数据稀缺问题，通过学生教师学习（KD）方法。
methods: 本研究使用的方法包括提议一种多语言学生教师（MUST）学习方法，利用一个预训练的映射模型将教师语言的 posterior 映射到学生语言的 ASR 模型中。
results: 根据实验结果，使用 MUST 学习方法可以将Relative Character Error Rate（CER）降低到9.5%，相比基eline monolingual ASR 模型。Abstract
Student-teacher learning or knowledge distillation (KD) has been previously used to address data scarcity issue for training of speech recognition (ASR) systems. However, a limitation of KD training is that the student model classes must be a proper or improper subset of the teacher model classes. It prevents distillation from even acoustically similar languages if the character sets are not same. In this work, the aforementioned limitation is addressed by proposing a MUltilingual Student-Teacher (MUST) learning which exploits a posteriors mapping approach. A pre-trained mapping model is used to map posteriors from a teacher language to the student language ASR. These mapped posteriors are used as soft labels for KD learning. Various teacher ensemble schemes are experimented to train an ASR model for low-resource languages. A model trained with MUST learning reduces relative character error rate (CER) up to 9.5% in comparison with a baseline monolingual ASR.
摘要
学生教师学习或知识蒸馏（KD）已经曾用于解决训练语音识别（ASR）系统的数据稀缺问题。然而，KD 训练的一个限制是学生模型类型必须是教师模型类型的正确或错误子集。这会防止训练不同语言的扩展，即使字符集不同。在这种情况下，我们提出了一种多语言学生教师（MUST）学习方法，利用 posterior mapping 技术。我们使用一个预训练的映射模型将教师语言的 posterior 映射到学生语言 ASR 中。这些映射 posterior 用作 KD 学习的软标签。我们对各种教师集合方案进行了实验，以训练一个低资源语言的 ASR 模型。与基线单语言 ASR 模型相比，我们的 MUST 学习方法可以降低相对字符错误率（CER）的差异为9.5%。
Counterfactually Probing Language Identity in Multilingual Models
results: 研究发现，给定一个 Language X 模板，向 Language Y 方向推动 embedding 会系统性地增加 Language Y 词汇的概率，超过第三方控制语言。但是，这并不特别地推动模型转化为翻译相当的 Language Y 词汇。向 Language X 方向推动也有一定的效果，但是会有些程度下降。总之，这些结果表明大量多语言语言模型具有both语言特定和语言通用的结构，并且 counterfactual probing 可以成功应用于多语言模型。Abstract
Techniques in causal analysis of language models illuminate how linguistic information is organized in LLMs. We use one such technique, AlterRep, a method of counterfactual probing, to explore the internal structure of multilingual models (mBERT and XLM-R). We train a linear classifier on a binary language identity task, to classify tokens between Language X and Language Y. Applying a counterfactual probing procedure, we use the classifier weights to project the embeddings into the null space and push the resulting embeddings either in the direction of Language X or Language Y. Then we evaluate on a masked language modeling task. We find that, given a template in Language X, pushing towards Language Y systematically increases the probability of Language Y words, above and beyond a third-party control language. But it does not specifically push the model towards translation-equivalent words in Language Y. Pushing towards Language X (the same direction as the template) has a minimal effect, but somewhat degrades these models. Overall, we take these results as further evidence of the rich structure of massive multilingual language models, which include both a language-specific and language-general component. And we show that counterfactual probing can be fruitfully applied to multilingual models.
摘要
使用 causal 分析技术可以探索语言模型（mBERT 和 XLM-R）中的语言信息结构。我们使用一种方法——Counterfactual probing，以探索这些模型的内部结构。我们在一个 binary 语言标识任务上训练了一个线性分类器，以分类Token是来自哪种语言。通过对这些分类器权重进行Counterfactual probing操作，我们可以将表示Vector проек到null空间中，并将其推动向Language X 或 Language Y 方向。然后，我们在一个隐藏语言模型任务上进行评估。我们发现，当给定一个 Language X 模板时，推动向 Language Y 方向会系统地增加 Language Y 词汇的概率，而这与第三种控制语言相比，这种效果明显。但是，不会 Specifically push 模型向翻译相同的 Language Y 词汇方向。推动向 Language X （与模板相同的方向）的效果相对较小，但是会有一定程度的降低这些模型的性能。总之，我们认为这些结果是证明大型多语言语言模型具有了rich结构，包括语言特定和语言通用的组成部分。同时，我们示出了对多语言模型的 counterfactual probing 可以得到有用的结果。